{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "scale": 1.3
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "scale": 0.7
        },
        "mean": 17.77549433708191
      },
      {
        "cfg": {
          "lr": 0.003,
          "scale": 1.0
        },
        "mean": 13.713647603988647
      },
      {
        "cfg": {
          "lr": 0.01,
          "scale": 1.3
        },
        "mean": 7.7490280866622925
      }
    ],
    "full": {
      "mean": 7.927547872066498,
      "std": 0.6638450060890482,
      "per_seed": [
        6.863780498504639,
        8.03250503540039,
        8.228401184082031,
        7.871425628662109,
        6.849928855895996,
        8.388378143310547,
        8.470486640930176,
        8.715476989746094
      ],
      "n": 8
    }
  },
  "idea": {
    "cfg": {
      "lr": 0.01,
      "scale": 1.3
    },
    "mean": 9.446731090545654,
    "std": 0.9494091152353562,
    "per_seed": [
      10.052746772766113,
      10.37270736694336,
      8.832763671875,
      8.52699089050293,
      7.573822021484375,
      10.16633129119873,
      10.212395668029785,
      9.836091041564941
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 1.5191832184791565,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      3.1889662742614746,
      2.3402023315429688,
      0.6043624877929688,
      0.6555652618408203,
      0.7238931655883789,
      1.7779531478881836,
      1.7419090270996094,
      1.1206140518188477
    ],
    "p_value": 0.0081,
    "mde": 0.7710830795135443,
    "mde_rel_pct": 9.726627854629957,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "track_choice": "tabular: initialization is explicitly covered by the benchmark mapping; same mlp_tiny and training loop used.",
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.01,
        "scale": 1.3
      }
    },
    {
      "cfg": {
        "lr": 0.001,
        "scale": 0.7
      }
    }
  ],
  "trained_geometry": {
    "baseline_median_nn": 0.4159104786813259,
    "idea_median_nn": 0.06890426948666573,
    "baseline_mean_nn": 0.4142976365983486,
    "idea_mean_nn": 0.07547109667211771
  },
  "mechanism_signature": {
    "quantity": "trained first-layer median nearest-neighbor distance",
    "prediction": "Coulomb repulsion preserves larger inter-row spacing as N=64 bank grows",
    "baseline_observed": 0.4159104786813259,
    "idea_observed": 0.06890426948666573,
    "observed_ratio_idea_over_baseline": 0.1656709148207366,
    "confirmed": false,
    "n_models": 8
  }
}